What it does
A Claude Code skill that improves the code around a fixed model, such as its memory, retrieval, context building, summaries, prompt templates and tool selection. It runs a loop that proposes changes, scores them and keeps the candidates that no other candidate beats on every score. It is a native version of the Meta-Harness method.
Use cases
- 01Test changes to retrieval or context building
- 02Compare prompt templates against the same scoring rules
- 03Keep a log of every scaffolding variant that was tried